Open weightsAlibaba Qwen/10 trustVerified

Qwen3-Coder

Name: Qwen3-Coder
Rating: 8

Alibaba's 480B open MoE built for agentic coding — 256K context, Apache 2.0

Visit official page →

About

Qwen3-Coder is the Qwen team's flagship open agentic-coding model, launched July 22, 2025 (a newer compact sibling, Qwen3-Coder-Next, is a separate model). The flagship checkpoint Qwen3-Coder-480B-A35B-Instruct is a 480B-parameter MoE with 35B active parameters, 256K native context (extensible to 1M with YaRN), trained on 7.5T tokens with 70% code across 358 languages. Alibaba pairs it with Qwen Code, an open-source terminal coding agent, and the model works with Claude Code, Cline, and other agentic harnesses. Released under Apache 2.0.

Specs

Context window

256K tokens native (1M with YaRN extrapolation)

Pricing

$0.30 input / $1.00 output per 1M tokens (DeepInfra, Oct 2026; varies by host)

Free quota

Free to download and self-host; hosted APIs charge per token (trial credits vary by provider)

API access

Alibaba Cloud Model Studio API and chat.qwen.ai; many third-party hosts (DeepInfra, Requesty, etc.); runs locally via Ollama, vLLM, llama.cpp

Weights

Open weights

Released

July 22, 2025

Benchmarks

SWE-Bench Verified: 69.6% — provider-reported (SOTA among open models at release)
Aider-Polyglot: 61.8%, TAU-bench Retail 77.5%, BFCL-v3 68.7% — third-party aggregator (llm-stats; methodology unclear)
MMLU Pro 78.8%, GPQA Diamond 61.8%, LiveCodeBench 58.5% — third-party aggregator (Requesty; methodology unclear)

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

codeautomationreasoningchattestingdevops

Trust score — 8/10

/10 trustVerified

Flagship identity, launch date (2025-07-22), Apache 2.0 license and specs confirmed against the Qwen launch blog and multiple independent guides. Minor caveats: model is 15 months old and superseded in spirit by the Qwen3.8 line; benchmark figures beyond SWE-Bench Verified come from third-party aggregators with unclear methodology. Red flags: none (note: Alibaba's newer Qwen3.8-Flash-Next carries a restrictive license for coding-assistant businesses — this repo is confirmed Apache 2.0, but always re-check the repo's LICENSE file, not the brand name) Metrics-v1 score 8/10 — adoption 8.9, maintenance 5, openness 10, manual 9.

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 8.9/10
Maintenance · recency, releases, issue load: 5.0/10
Openness · license + docs: 10.0/10
Manual review · sanity read + red-flag check: 9.0/10

· HF top match: Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8
· HF downloads 583,513 (peer-normalized)
· released 448 days ago
· manual review: 15 months old; non-SWE-Bench figures from aggregators with unclear methodology